My thesis is that, with such short prompts, the user is doomed to receive a milquetoast average response; the bot correctly identifies the specific domain but elaborates a global non-specific approach that isn’t sufficiently nuanced.

Seems you’re on to something - I submitted the phrase “Demoknight TF2” (an old(ish) meme started by/referencing TF2ber SolarLight, and referencing the Demoknight subclass, which explicitly forgoes a Stickybomb Launcher for a shield), and it gave me a godawful slop guide to the aforementioned subclass:

Going over a few things obviously wrong with this:

Demoknight is a popular TF2 “Demoman” build/theme

w h a t ?

mobile melee/ranged threat

Hybrid Demoknight (which takes a grenade launcher in the primary slot), I can see describing as a ranged threat. Full Demoknight (which takes Ali Baba’s Wee Booties instead), not so much.

usually with a shield for mobility/survivability and a sword or axe for damage

Demoknight explicitly forgoes the Stickybomb Launcher for a shield of some description

Scotsman (sword/knight vibe)

There’s no weapon called the Scotsman, but there is an axe called the Scotsman’s Skullcutter

or other sword/axe-style options depending on the playstyle

There’s only one axe for the Demo in the vanilla game - the aforementioned Skullcutter.

Shield: Tide Turner (more aggressive with better survivability/tempo) or Splendid Screen (more defensive, survivability-oriented).

There is a survivability-oriented shield, but its neither of these - its the Chargin’ Targe, which offers 50% fire resist and 30% explosive resist (the Screen offers 30% resist on both, and the Turner offers only 15% on both, for comparison.

Calling the Splendid Screen “defensive” is also pretty far off the mark - beyond the aforementioned reduced resistances, its main advantage is a faster recharge on the shield charge, and a 70% increase to shield bash damage

As for the Tide Turner, it trades in offensive capabilities (it prevents its user from landing charge crits with their melee) for mobility-related ones (it completely disables turn control restrictions).

Grenade / support: often a grenade launcher for area control when you can’t reach melee cleanly, or a weapon that complements your survivability/engagements.

Yeah, seems the chatbot can’t tell the difference between full and hybrid knight. Also, the grenade launchers are generally better for landing direct hits - area control is more the domain of a stickybomb launcher, or the Soldier’s Rocket Launcher(s).

Use explosives smartly: not constant spam—more like zoning, forcing movement, or punishing groups.

Seems the chatbot got Demoknight and stock Demo mixed up.

Demoknight in general should stump chatbots pretty easily - beyond being effectively a different class from stock Demo, the subclass has a lot of nuance that can and will trip up the chatbots. To quote SolarLight’s hour-long video on the charge mechanic:

“Well, at least you know that all of Demoknight’s weapons have their uses when it comes to charging, and that Demoknight’s charge can be essentially be boiled down into three different types (listed on-screen as “Full Control”, “Some Control”, and “Low Control”). There’s also a distinction between Full Demoknight (melee-only) and Hybrid Demoknight (melee + grenade launcher), which is often a massive shift in playstyle. Demoknight has so much​ ​weapon variety, that talking about specific strategies can be a bit of a nightmare. Any given strategy might be godlike in one loadout, suboptimal in another, and impossible in another.”

OT: What the fuck happened whilst I was asleep

alt: A screenshot showing nineteen fucking notifs

On one hand, this is awful. On the other hand, if I worked at a garbage tech company and my boss would not only pay me to sabotage the code base and spend their money but also reward me for doing so… I mean that’s an attractive proposition.

Before, sabotaging your competition was difficult and legally risky. AI fixes this - now, your competition shall sabotage themselves, leaving you free to focus on other things.

The Carlson Lab has agreed on blanket-banning chatbots for writing. Why? Simple:

Using generative AI inherently exposes you to the risk of career-ending accusations of plagiarism.

In a reasonable world, this would have probably sent everyone running for the hills. Moral issues of AI aside, a fancy way of committing career suicide is never worth whatever upsides it offers. We do not leave in a reasonable world.

Tamagotchi didn’t die for this

Hank Green is a popular science YouTuber who’s been pushing AI doomsday harder and harder for about the past year. The Washington Post article details some of the sponsorships that paid Green to shill for the AI doomsday cult — even as Green says in the article that he thinks AI doomsday is silly.

Green has personal experience of AI doom — when it came out last week just how much of his stuff was written with ChatGPT and Claude. [Mashable]

Hank stood too close to the circus, and caught a stray as a result. You love to see it.

I found the song - was a fun listen all around.

Somehow managed to find a sneer on LinkedIn, of all fucking places:

“It’s open source, so anyone can verify” – unfortunately for them the verifier was also a hacker.

Also, most of the would-be verifiers are frying their brains with AI

Gamers Nexus has announced plans to poison the LLMs scraping them, alongside updates to the megacharts on their website.

The Silicon Valley set’s also shown a complete inability to tell good work from slop, as their relentless glazing of AI slop has demonstrated, so I can easily believe that.

You hide your prompt in a document, which can even be completely outside the target’s Copilot organisation. That infected document gets used as a source by Copilot for Word. If Copilot sees your prompt and uses it as instructions, it may manipulate the document the user is editing — change financial numbers, send company data out to the attacker, and so on.

The new document may in turn become a carrier and spread the infection — even without the original infected document being present. You now have an AI worm.

Oh, joy, we’ve been overdue for another ILOVEYOU-level incident.

A new way to fuck with AI scrapers just dropped!

Aaaaaand its fucking with screen readers, making protected text inaccessible to blind readers. File this next to the alt-text drama in the “AI bros set back accessibility” bucket.

(from the youtube comments)

shot:

One of the lesser discussed issues with the generative AI hype is that it put every single deep learning application under scrutiny. They tend to hide behind the positive uses of deep learning whenever generative AI is criticized, as if medicine research and the misinformation generator cannot be separated. “Nuclear power is impossible unless we are allowed to use nukes” type of argument

chaser:

Alphafold was a really interest project. It was not magic but had some utility. Now all remaining AI projects are lying plagiarism machines.

AI is officially a pseudoscience now.

LLM bros actively force themselves everywhere, part infinity:

Newgrounds moderation had to deal with someone (probably a scammer) posting slop “art advice” recently - exactly why, nobody fuckin’ knows:

A lot of annoying people in technical areas view the world in terms of an intelligence hierarchy: the smartest people do math and physics, the slightly less smart people do coding, and the dumb people do everything else. So if AI can do math then it can do anything else. But, as an example, it is abundantly obvious now that AI is not replacing filmmaking.

Going by those annoying peoples’ logic, filmmakers are smarter than coders, because LLMs can (allegedly) program, but they can’t make a good film. I have no wider point to this, I just find this really, really funny

I tested this with the Navy Seal Copypasta (well, the first 400 letters of it, the demo’s got a limit) - using just Cryllic characters got me a 3.01x increase, and turning on all four control types got a 4.53x increase.

Testing your own recent comment, Cryllic only got 3.38x, and all four controls got 5.11x.

Going from those two, the boost from homoglyphs alone is likely higher than you think - 3x to 3.5x, by my guess - pretty good for human readable text.

Tarpits like Iocaine and Nepenthes can easily sacrifice readability for token burn, so they can easily go higher.

Adversarial tokenmaxxing could probably be done by using non-English character sets in lieu of English letters (e.g. faux Cryllic) - for two examples from the Greek alphabet, alpha and omicron alone can easily substitute for A and O, respectively.

As a bonus, this would likely make the text look like complete gibberish to LLMs, potentially leaving them unable to process the document altogether. This would probably shaft anyone using screen readers, though.

EDIT: Turns out the demonstration’s already caught on to this idea, didn’t notice beforehand:

New(ish) Baldur Bjarnason - a fairly politically charged one at that, going into the US hegemony powering the current tech industry (and the AI bubble by extension), and how the Hormuz crisis is all-but guaranteed to topple the whole thing.

AI is bad at everything, part infinity: AI transcription whitewashes 18th-century documents